We prospectively enrolled children <18 years of age who were hospitalized because of clinical and radiographic pneumonia during April 2017‒January 2020 at Chang Gung Memorial Hospital Lin Kou Branch (Taoyuan, Taiwan), Chang Gung Memorial Hospital Kaohsiung Branch (Kaohsiung, Taiwan), and Saint Paul’s Hospital (Taoyuan, Taiwan). We obtained written informed consent from patients or their parents. Throat swab specimens were then collected by pediatricians who used sterile swabs (FLOQSwabs; Copan, https://www.copanusa.com). All throat swab specimens were sent to a laboratory at the Chang Gung Memorial Hospital Lin Kou Branch. We inoculated throat swab specimens into SP4 medium and observed spherical M. pneumoniae colonies by using microscopy. This study protocol was approved by the research ethics committee of Chang Gung Memorial Hospital, Taiwan (approval nos. 201900420A3 and 202000687B0).

We obtained whole-genome sequencing (WGS) data for 99 clinical isolates. We cultured M. pneumoniae in 75-cm² tissue culture flasks (Techno Plastic Products AG, https://www.tpp.ch) containing 10 mL of SP4 medium for 96 hours at 37°C. We extracted M. pneumoniae genomic DNA directly from cultures by using the QIAamp DNA Mini Kit (QIAGEN, https://www.qiagen.com), and prepared libraries for WGS by using NEBNext Ultra II DNA Library Prep Kits (Illumina, https://www.illumina.com).

We combined 185 isolates that had available WGS data, downloaded from the RefSeq database of the National Center for Biotechnology Information (NCBI; https://www.ncbi.nlm.nih.gov), and our assembled 99 clinical isolates (Appendix 1 Table). We used all 284 M. pneumoniae isolates to analyze homologous recombination and genetic diversity as described (10).

We used Gubbins (https://sanger-pathogens.github.io/gubbins) to identify areas that were likely introduced by homologous recombination involving M. pneumoniae clades. We scored each gene with the number of predicted recombination events in each clade, and used R version 3.6.3 (R Foundation for Statistical Computing, https://cran.r-project.org) to draw a heatmap by using the pheatmap package.

We annotated 284 M. pneumoniae by using Prokka version 1.14.6 (11) and identified core genes by using Roary version 3.13.0 (12) with default settings. We aligned sequences of genes by using ClustalW2 (13) and converted protein and DNA multiple sequence alignments into codon alignments by using PAL2NAL (14). We used the output data file to estimate synonymous and nonsynonymous substitution rates (K_a/K_s) in different branches of the phylogenetic tree by using a branch-specific algorithm in the codeml program from PAML version 4.8 (15), for which we provide additional details (Appendix 2). We used M129 as a reference genome to describe the location of RepMPs based on the NCBI annotation file.

We used whole-genome alignments to construct a phylogenetic tree by using RAxML (https://cme.h-its.org/exelixis/software.html) with the GTRCAT model. We performed root-to-tip regression analysis by using TempEst version 1.5.3 (16). To estimate the phylodynamics of M. pneumoniae, we performed temporal analysis using the BEAST version 1.10.4 software package (17). We used a strict clock exponential model and a Hasegawa–Kishono–Yano nucleotide substitution model with a discretized gamma distribution to measure rate heterogeneity across sites. We summarized tree data to generate a maximum clade credibility tree by using TreeAnnotator (https://beast.community/treeannotator#user-interface) and visualized it by using FigTree version 1.4.4 (1).

We used PLINK version 1.9 (18) to generate a binary output from the vcf file. We performed population structure inference by using ADMIXTURE version 1.3.0 (19).

From the NCBI database, we collected M. pneumoniae strains from China (n = 23), Japan (n = 70), South Korea (n = 30), the United States (n = 29), the United Kingdom (n = 4), Denmark (n = 3), Spain (n = 1), France (n = 15), Germany (n = 2), Guatemala (n = 1), Egypt (n = 2), Kenya (n = 3), and Tunisia (n = 2) obtained during 1944‒2016 (Figure). These strains included P1 subtype 1 sequence type (ST) 3, prevalent worldwide; P1 subtype 2 ST2 from Japan, the United States, and countries in Europe; P1 subtype 1 ST1 from the United States, South Korea, China, and Tunisia; P1 subtype 1 ST17 from South Korea; P1 subtype 2 ST7 from Japan; and P1 subtype 2 ST14 from South Korea, Japan, France, and the United States.

We observed macrolide resistance with the A2063G mutation among ST3 strains from China, South Korea, Japan, and France; ST1 strains from China; ST7 strains from Japan; and ST14 strains from South Korea and Japan. A total of 99 strains from Taiwan collected during 2017‒2020 consisted of P1 subtype 1 ST3 (n = 60, 60.6%), P1 subtype 1 S17 (n = 31, 31.3%), P1 subtype 2 ST14 (n = 7, 7.1%), and P1 subtype 2 ST2 (n = 1, 1%). Macrolide resistance with A2063G was also observed in ST3 (n = 59), ST17 (n = 14), and ST14 (n = 3). A2063T was observed in ST17 (n = 10), A2064G in ST3 (n = 1).

We inferred genomic regions identified by Gubbins as having densities of single-nucleotide variants (SNVs) that were different from background SNV densities in the core genome to be homologous recombination events. By examining all 284 M. pneumoniae genomes in this study, we detected 108 putative recombination blocks (Appendix 2 Figure 1, panel A) spanning an average of 1.3 kb/recombination event (95% CI 0.9–1.7 kb) and covering an average of 10 kb/isolate (1.3% of the genome).

Phylogenetic analysis using core genome single-nucleotide polymorphisms (SNPs) identified 5 clades linked to multilocus sequence typing structures and associated with P1 subtypes, which resembled previous findings (1,7,20). Clade T1–1 consisted of ST1; clade T1–2 consisted of ST3, ST19, ST20, and ST30; clade T1–3 consisted of ST17; clade T2–1 consisted of ST2, ST4, ST7, and ST16; clade T2–2 consisted of ST2, ST14, ST15, and ST33. Clades T1–1, T1–2, and T1–3 belonged to P1 subtype 1, and clades T2–1 and T2–2 belonged to P1 subtype 2.

With regard to switching from subtype 1 to subtype 2, two regions of SNPs were predicted to have arisen by homologous recombination: first, 183717–186553 bp sequence coordinates containing P1 adhesin (MPN141) for T2–1 and 183381–187741 bp for T2–2; and second, 437080–443621 bp sequence coordinates containing hypothetical proteins (MPN366–367), RepMP1, DUF16 domain-containing protein (MPN368), lipoprotein (MPN369), putative adhesin (MPN370), and MPN371 for both clades T2–1 and T2–2. Among strains of subtype 1, clade T1–2 had 3 regions of SNPs predicted to have arisen by homologous recombination: first 34,0697–340757 bp sequence coordinates containing prrB (MPN285); second, 493781–493791 bp sequence coordinates containing adhesin (MPN409); and third, 570767–570790 bp sequence coordinates containing adhesin (MPN468) (Appendix 1 Table 2; Appendix 2 Figure 1, panel A).

The MPN409 recombined region was more frequently detected in most macrolide-resistant clade T1–2 strains than in macrolide-susceptible clade T1–2 strains. We ran Gubbins independently for subtype 2 (88 isolates) based on the core genome sequence alignment involving M. pneumoniae subtype 2 reference strain FH (GenBank accession no. CP010546.1). Among strains of subtype 2, only 1 recombination hotspot was predicted: 179477–183466 bp sequence coordinates containing P1 adhesin (equal to MPN141 of M129) (Appendix 1 Table 2; Appendix 2 Figure 1, panel B).

We estimated 2 measures of the recombination rate in the core genome alignment by using Gubbins ρ/θ and r/m ratios (Table). We provide boxplots of putative recombination statistics for M. pneumoniae clades (Appendix 2 Figure 2). ρ/θ estimates the relative frequency of occurrence of recombination and mutation in the history of the clade. The overall ρ/θ ratio for each clade ranged from 0 (clade T2–1) to 0.03 (clade T1–2), indicating that putative recombination events have occurred less frequently than vertically inherited mutations in all 5 clades, and lower than in most microbes. The r/m index assesses the relative effect of recombination and mutation on genetic diversification of the clade. Predicted recombination occurred more frequently in clade T1–2, for which the r/m rate was 4–13 times higher than for the other 4 clades.

We defined clade-specific genes involved in these putative recombination events by clustering of genes by recombination frequency in each clade (Appendix 2 Figure 3, panel A). These data can guide identification of multiple genes under different selective pressures in each clade. A set of 8 genes (MPN141, MPN142, MPN366, MNP368, MPN369, MPN370, MPN_RS02085, and MPN_RS02055) appeared to increase the recombination frequency in M. pneumoniae subtype 2, and extensive DNA repetitive elements are located in each of these genes. MPN366, MPN368, MPN369, and MPN370 were under even greater selective pressure than MPN141 and MPN142 (Appendix 2 Figure 3, panel A).

We investigated adhesin P1 sequence variation by calculating homology with the reference M129 strain DNA sequence (Appendix 2 Figure 3, panel B). Two variable regions provided most variation between M. pneumoniae subtypes 1 and 2, and were located at ≈3000 bp and ≈1000 bp within MPN141. In contrast, these 2 regions were highly conserved within clades T1–1, T1–2, and T1–3. For subtype 2, clade T2–2 had less conserved sequence than clade T2–1.

Two additional regions of nucleotide sequence differed among strains belonging to clade T2–2 compared with M129. Sequence analysis of the P1 gene showed divergent sequences from 5 clades at position 2000 bp, which might be an AGT trinucleotide variable-number tandem repeat (21).

To clarify this potentially confounding within‐clade mutation, we analyzed the genetic diversity of the collection by estimating the nonsynonymous to synonymous (K_a/K_s) mutation ratio. We assessed divergence along the genome between clades by determining synonymous substitution rates, a near-neutral indicator of genetic divergence. Data showed that different genomic regions had higher divergence among those 5 clades. Clade T2–1 had more divergent regions than clade T2–2, and clade T2–2 had higher K_s values in less divergent regions. The genomic divergence pattern of clade T1–2 was observed across the genome. This result indicated that clade T1–2 was the most divergent in M. pneumoniae subtype 1, and the genome of clade T1–3 was stable without evidence of divergence (Appendix 2 Figure 4, panel A). We also identified numerous diverse regions located near RepMP elements across the genome. We further investigated the function of these genomic regions that showed higher divergence. From a global perspective, numerous positions are located in genes that are annotated as adhesin and MgPa (mgp) operon of M. genitalium coding for proteins (Appendix 2 Figure 4, panel B).

To track genomic evolutionary footprints, we used codeml analysis to test genes for evidence of positive selection. We detected 42 genes as being under positive selection by using likelihood ratio tests (Appendix 1 Table 3). Of these positively selected genes, 21 were core orthologous loci present among all 284 genomes investigated. To determine the relationship of genes with certain evolutionary features, we classified these genes into different functional categories by using clusters of orthologous groups of proteins annotation. There are a large number of genes in the functional categories translation, ribosomal structure, and biogenesis and cell wall/membrane/envelope biogenesis. The cell wall/membrane/envelope biogenesis category includes membrane-associated genes under positive selection, consistent with observations for other bacterial species (22). The translation, ribosomal structure, and biogenesis category, including ribosomal biogenesis genes, might be correlated with drug resistance.

We calculated the temporal history of M. pneumoniae by using Bayesian evolutionary analysis sampling trees to place the most recent common ancestor in ≈1300 (95% highest posterior density 1230–1392) by exponential growth analysis (Appendix 2 Figure 5). The maximum clade credibility tree also showed 2 distinct clades that are consistent with the 2 subtypes for P1 gene variation. This analysis suggested that the 2 subtypes might have diverged at approximately the same time as the most recent common ancestor arose. The evolutionary divergence time estimates for subtypes 1 and 2 for the 5 clades were from 1800 to 1900 (subtype 1, 1823–1865; subtype 2, 1811–1855). M. pneumoniae subtypes 1 and 2 spread rapidly within 20 years (through the year 2000).

Genetic history can be established by examining the patterns of shared genetic variation between >2 populations. Admixture proportion inference is an algorithm used to infer the proportions of ancestry from each source population based on shared genetic drift. These proportions can produce visual summaries that identify a population structure. We used ADMIXTURE to fit a model which the genome is composed of sites from K = 2 to K = 10 (K = n, where n is the number of ancestral generations) for ancestral populations. We provide ADMIXTURE analysis with estimated proportions of 7 ancestral populations (Appendix 2 Figure 6).

We provide unsupervised models assuming K = 4 and K = 7 ancestral clusters (Appendix 2 Figure 6, panel A). At K = 7, it is difficult to discern the major contributors to each clade, which is likely to result in a disrupted relationship between clades T2–1 and T1–1 and other clades. Because each column represents 1 isolate, we show data for admixed strains observed particularly in clades T1–2 and T2–2, reflecting the divergence of the 2 clades (Appendix 2 Figure 6, panel A).

To further clarify the genetic track, we mapped populations to geographic region (Appendix 2 Figure 6, panel B). Each pie chart shows the mean composition of 7 populations in each region. The distribution shows genetic diversity related to geography. Thus, M. pneumoniae has the potential to be transmitted efficiently and spread regionally from Europe to America, and subsequently emerge rapidly in Asia. Genome sequences generated during this study were deposited into the NCBI BioProject database (accession no. PRJNA699672).